Chapter V · Limits of functions of a continuous variable. Continuous and discontinuous functions

Implicit and inverse functions

§§109–110 35 min

When an equation f(x,y)=0f(x, y) = 0 really defines yy as a continuous function of xx, and why a continuous strictly monotone function has a continuous inverse.

Why this matters

In Chapter II Hardy spoke freely of “the function yy defined by y5−xy−y−x=0y^5 - xy - y - x = 0”, and of x3\sqrt[3]{x}, arcsin⁡x\arcsin x and their like, without asking whether such functions exist. An equation may have no solution yy for some xx, or several. In §§109–110 he settles the question near a point: if ff is continuous and strictly monotone in yy, the equation f(x,y)=0f(x, y) = 0 defines exactly one continuous function y=ϕ(x)y = \phi(x) through the given point. The special case f(x,y)=F(y)−xf(x, y) = F(y) - x gives inverse functions, which we need for xn\sqrt[n]{x}, arcsin⁡x\arcsin x, and above all for the logarithm as the inverse of the exponential (Chapter IX).

The idea

Hardy’s argument is a picture (his Fig. 33). Suppose f(a,b)=0f(a, b) = 0 and that, in a square round P=(a,b)P = (a, b), ff is continuous and strictly increasing in yy for each fixed xx.

  • Go straight up from PP to a point Q=(a,b+k)Q = (a, b + k) and straight down to R=(a,b−k)R = (a, b - k). Since ff increases in yy, f(Q)>0>f(R)f(Q) > 0 > f(R).
  • By continuity these signs persist a little to the left and right: f(x,b+k)>0f(x, b + k) > 0 and f(x,b−k)<0f(x, b - k) < 0 for all xx near aa.
  • On each such vertical line, y↦f(x,y)y \mapsto f(x, y) is continuous and goes from negative to positive, so it vanishes somewhere (the intermediate value theorem), and being strictly increasing it vanishes only once. Call that point ϕ(x)\phi(x).
Explore Hardy’s construction

Try this. Drag the purple handle to move the vertical line x=x1x = x_1 across the square, and watch the slice y↦f(x1,y)y \mapsto f(x_1, y) on the right. Note where the signs at QQ and RR stay opposite (the dashed band ∣x∣≤h|x| \le h). Then choose y2−x=0y^2 - x = 0 and do the same.

Solving f(x,y)=y5−xy−y−x=0f(x, y) = y^5 - xy - y - x = 0 for yy near (0,0)(0, 0)
xy
yf(0.15, y)
  • f > 0
  • f < 0
  • f = 0
  • square |x|, |y| ≤ k; dashed: |x| ≤ h
  • the line x = x₁ (drag the handle) and y ↦ f(x₁, y)
At x1=0.15x_1 = 0.15: f(Q)=f(x1,0.4)=−0.6f(Q) = f(x_1, 0.4) = -0.6, f(R)=0.3f(R) = 0.3 (opposite signs); the segment QRQR contains 1 solution (−0.1305-0.1305). Here ff is strictly decreasing in yy throughout the square (for small kk), so whenever f(Q)f(Q) and f(R)f(R) differ in sign there is exactly one solution; that holds for ∣x∣≤h≈0.278|x| \le h \approx 0.278.
What to notice

For the first two equations the slice is a strictly decreasing curve, so it crosses 00 exactly once while f(Q)f(Q) and f(R)f(R) differ in sign; the crossings trace out the continuous curve y=ϕ(x)y = \phi(x). For y2−xy^2 - x the slice is a U-shape: two crossings when x>0x > 0, none when x<0x < 0. Monotonicity in yy is exactly what fails, and with it uniqueness and existence.

Precise statements

Hardy uses the word neighbourhood: a condition holds “in the neighbourhood of (a,b)(a, b)” if it holds throughout some square centred at (a,b)(a, b), and “in the neighbourhood of aa” if it holds in some interval round aa.

Theorem The implicit function theorem Hardy: §109

Suppose that, in the neighbourhood of (a,b)(a, b),

  1. f(x,y)f(x, y) is a continuous function of xx and yy;
  2. f(a,b)=0f(a, b) = 0;
  3. for each xx near aa, f(x,y)f(x, y) is a strictly increasing function of yy (or strictly decreasing).

Then, in the neighbourhood of aa, there is a unique function y=ϕ(x)y = \phi(x) with values near bb (in the square) such that f(x,ϕ(x))=0f(x, \phi(x)) = 0 identically; it satisfies ϕ(a)=b\phi(a) = b, and it is continuous in the neighbourhood of aa.

Hardy points out that the neighbourhood in the conclusion is generally smaller than the square in the hypotheses (in the widget, hh is usually less than kk).

Taking f(x,y)=F(y)−xf(x, y) = F(y) - x, which is strictly monotone in yy exactly when FF is, gives a local inverse; Hardy then uses Heine–Borel to make it global, and finally gives a direct proof.

Theorem Inverse functions Hardy: §110

If F(y)F(y) is continuous and strictly increasing for a≤y≤ba \le y \le b, with F(a)=AF(a) = A and F(b)=BF(b) = B, then there is a unique function y=ϕ(x)y = \phi(x) on [A,B][A, B] with F(ϕ(x))=xF(\phi(x)) = x; it is continuous and strictly increasing from aa to bb. The same holds with “decreasing” throughout.

Geometrically, the graph of ϕ\phi is the graph of FF reflected in the line y=xy = x.

Explore The inverse as a reflection

Try this. For x=y3x = y^3, drag the purple point along the curve and watch its mirror image. Then choose x=y2x = y^2 and x=y5−yx = y^5 - y, and change the interval [a,b][a, b] until the reflected curve becomes the graph of a function.

x=F(y)=y3x = F(y) = y^3 for −1.2≤y≤1.2-1.2 \le y \le 1.2, and its reflection
  • the curve (y, F(y)), drawn with y across
  • its reflection (F(y), y): the inverse, if it is a graph
  • the line of reflection
  • the level x = F(y) and its solutions
F is strictly increasing on [−1.2,1.2][-1.2, 1.2], so every level xx between −1.73-1.73 and 1.731.73 has exactly one solution yy, and the reflected curve is the graph of a continuous increasing inverse, y=x3y = \sqrt[3]{x}. The purple point (0.5,0.125)(0.5, 0.125) reflects to (0.125,0.5)(0.125, 0.5).
What to notice

Where FF is strictly monotone the reflected curve passes the vertical-line test: each xx has one yy, and the inverse is continuous. y2y^2 is not monotone on any interval round 00; restrict to [0,1][0, 1] and you get x\sqrt{x}, to [−1,0][-1, 0] and you get −x-\sqrt{x}. y5−yy^5 - y is decreasing on [−0.6,0.6][-0.6, 0.6], giving the unique solution of y5−y−x=0y^5 - y - x = 0 that vanishes with xx; on a wider interval some levels xx are taken three times.

Proofs

Proof The implicit function theorem
Strategy

Build ϕ\phi by Hardy’s construction (existence and uniqueness on each vertical line), then get continuity by running the same construction in a smaller square: if the square has half-side ε\varepsilon, the solution is trapped within ε\varepsilon of bb.

Signs at the top and bottom

Take ff increasing in yy and a square ∣x−a∣,∣y−b∣≤k|x - a|, |y - b| \le k where the hypotheses hold. Strict monotonicity in yy at x=ax = a gives f(a,b+k)>f(a,b)=0>f(a,b−k)f(a, b + k) > f(a, b) = 0 > f(a, b - k).

The signs persist sideways

ff is continuous at (a,b+k)(a, b + k) and (a,b−k)(a, b - k), and x↦f(x,b±k)x \mapsto f(x, b \pm k) is continuous in xx (continuity in both variables implies it). By persistence of sign (§102) there is h≤kh \le k with f(x,b+k)>0f(x, b + k) > 0 and f(x,b−k)<0f(x, b - k) < 0 for ∣x−a∣≤h|x - a| \le h.

Exactly one zero on each vertical

Fix xx with ∣x−a∣≤h|x - a| \le h. The function y↦f(x,y)y \mapsto f(x, y) is continuous on [b−k,b+k][b - k, b + k], negative at the bottom and positive at the top, so by §101 it vanishes at some yy; being strictly increasing, it vanishes only once. Call that yy value ϕ(x)\phi(x). Then ϕ(a)=b\phi(a) = b, f(x,ϕ(x))=0f(x, \phi(x)) = 0, and no other yy in [b−k,b+k][b - k, b + k] works.

Continuity at aa

Given ε\varepsilon with 0<ε<k0 < \varepsilon < k, repeat the first two steps with ε\varepsilon in place of kk: there is hε>0h_\varepsilon > 0 with f(x,b+ε)>0>f(x,b−ε)f(x, b + \varepsilon) > 0 > f(x, b - \varepsilon) for ∣x−a∣≤hε|x - a| \le h_\varepsilon. For such xx the unique zero in [b−k,b+k][b - k, b + k] lies in (b−ε,b+ε)(b - \varepsilon, b + \varepsilon). So ∣ϕ(x)−b∣<ε|\phi(x) - b| < \varepsilon when ∣x−a∣≤hε|x - a| \le h_\varepsilon: ϕ\phi is continuous at aa.

Continuity at nearby points

For ∣x0−a∣<h|x_0 - a| < h, the point (x0,ϕ(x0))(x_0, \phi(x_0)) satisfies the same hypotheses (in a smaller square inside the first), so the same argument shows ϕ\phi is continuous at x0x_0. Hardy reaches the same conclusion through the upper and lower limits Λ\Lambda, λ\lambda of ϕ\phi at aa: both satisfy f(a,⋅)=0f(a, \cdot) = 0, so both equal bb.

Proof Inverse functions (Hardy’s direct proof)
Strategy

For a level ξ\xi between AA and BB, the solution of F(y)=ξF(y) = \xi should be the last yy at which FF is still at most ξ\xi: take η\eta to be the upper bound of {y:F(y)≤ξ}\{y : F(y) \le \xi\}. Then show F(η)=ξF(\eta) = \xi, uniqueness, monotonicity, and continuity.

Define η\eta

Let A<ξ<BA < \xi < B and S={y∈[a,b]:F(y)≤ξ}S = \{y \in [a, b] : F(y) \le \xi\}. SS contains aa and is bounded by bb, so it has an upper bound η\eta.

F(η)=ξF(\eta) = \xi

If F(η)>ξF(\eta) > \xi, then by continuity F>ξF > \xi on an interval [η−h,η][\eta - h, \eta], so η−h\eta - h would already be an upper bound of SS (no yy there or beyond has F(y)≤ξF(y) \le \xi, since FF increases): contradiction. If F(η)<ξF(\eta) < \xi, then by continuity F<ξF < \xi on some [η,η+h][\eta, \eta + h], so SS contains points beyond η\eta: contradiction. (The end values ξ=A,B\xi = A, B give η=a,b\eta = a, b.)

Uniqueness and monotonicity

If F(y1)=F(y2)F(y_1) = F(y_2) then y1=y2y_1 = y_2, since FF is strictly increasing. Write ϕ(ξ)=η\phi(\xi) = \eta. If ξ1<ξ2\xi_1 < \xi_2 then ϕ(ξ1)<ϕ(ξ2)\phi(\xi_1) < \phi(\xi_2): otherwise ϕ(ξ1)≥ϕ(ξ2)\phi(\xi_1) \ge \phi(\xi_2) would give ξ1=F(ϕ(ξ1))≥F(ϕ(ξ2))=ξ2\xi_1 = F(\phi(\xi_1)) \ge F(\phi(\xi_2)) = \xi_2.

Continuity

ϕ\phi is strictly increasing on [A,B][A, B] and its values fill the whole interval [a,b][a, b] (each yy in [a,b][a, b] is ϕ(F(y))\phi(F(y))). By the theorem at the end of §101, a monotone function that takes every value between its end values is continuous.

Worked examples and exercises

Exercise Hardy’s example ●○○ Hardy: §109

Show that f(x,y)=y5−xy−y−xf(x, y) = y^5 - xy - y - x satisfies the hypotheses of the implicit function theorem at (0,0)(0, 0), with “decreasing” in condition 3.

Hint
Factor f(x,y)−f(x,y′)f(x, y) - f(x, y').

Solution

ff is a polynomial, so it is continuous, and f(0,0)=0f(0, 0) = 0. For condition 3,

f(x,y)−f(x,y′)=(y−y′) (y4+y3y′+y2y′2+yy′3+y′4−x−1).f(x, y) - f(x, y') = (y - y')\,(y^4 + y^3y' + y^2y'^2 + yy'^3 + y'^4 - x - 1).

If ∣x∣,∣y∣,∣y′∣≤12|x|, |y|, |y'| \le \tfrac12, the bracket is at most 5⋅116+12−1<05 \cdot \tfrac1{16} + \tfrac12 - 1 < 0, so f(x,y)−f(x,y′)f(x, y) - f(x, y') has the sign opposite to y−y′y - y': ff is strictly decreasing in yy. Hence exactly one continuous function y=ϕ(x)y = \phi(x) satisfies the equation identically near x=0x = 0 and vanishes at x=0x = 0.

Exercise A quadratic, solved explicitly ●○○ Hardy: §109

For y2−xy−y−x=0y^2 - xy - y - x = 0, check the hypotheses at (0,0)(0, 0) and show that the function given by the theorem is ϕ(x)=12{1+x−1+6x+x2}\phi(x) = \tfrac12\{1 + x - \sqrt{1 + 6x + x^2}\}. Why is the other root not the answer?

Hint
f(x,y)−f(x,y′)=(y−y′)(y+y′−x−1)f(x, y) - f(x, y') = (y - y')(y + y' - x - 1).

Solution

f(x,y)−f(x,y′)=(y−y′)(y+y′−x−1)f(x, y) - f(x, y') = (y - y')(y + y' - x - 1), and the second factor is negative when x,y,y′x, y, y' are small, so ff is strictly decreasing in yy near (0,0)(0, 0). As a quadratic in yy, y2−(1+x)y−x=0y^2 - (1 + x)y - x = 0 has roots 12{1+x±(1+x)2+4x}=12{1+x±1+6x+x2}\tfrac12\{1 + x \pm \sqrt{(1 + x)^2 + 4x}\} = \tfrac12\{1 + x \pm \sqrt{1 + 6x + x^2}\}. At x=0x = 0 these are 00 and 11. The theorem’s ϕ\phi vanishes at 00 and is continuous, so it is the root with the minus sign (with the positive square root); the other root is near 11, outside any small square round (0,0)(0, 0).

Exercise Even and odd roots ●○○ Hardy: §110

Discuss the equations y2n+1=xy^{2n+1} = x and y2n=xy^{2n} = x (nn a positive integer) near x=0x = 0, y=0y = 0, in the light of §110.

Hint
Is y2n+1y^{2n+1} strictly increasing? Is y2ny^{2n} monotone on any interval containing 00?

Solution

y2n+1y^{2n+1} is continuous and strictly increasing on the whole line (if y<y′y < y' then y′2n+1−y2n+1y'^{2n+1} - y^{2n+1} has the sign of y′−yy' - y, as is clear when y,y′y, y' have the same sign and trivial otherwise). So x=y2n+1x = y^{2n+1} has a unique continuous inverse, y=x2n+1y = \sqrt[2n+1]{x}, defined for all xx and vanishing at 00.

y2ny^{2n} decreases for y<0y < 0 and increases for y>0y > 0, so it is monotone on no interval round 00. Correspondingly y2n=xy^{2n} = x has two solutions ±x2n\pm\sqrt[2n]{x} for x>0x > 0 and none for x<0x < 0. Restricting to y≥0y \ge 0 (where y2ny^{2n} is strictly increasing) gives the inverse x2n\sqrt[2n]{x} on [0,∞)[0, \infty).

Exercise ●○○ Hardy: Misc. Examples V, 20

As xx increases from −12π-\tfrac12\pi to 12π\tfrac12\pi, sin⁡x\sin x is continuous and strictly increasing from −1-1 to 11. Deduce the existence of a continuous, strictly increasing function arcsin⁡y\arcsin y on [−1,1][-1, 1] with sin⁡(arcsin⁡y)=y\sin(\arcsin y) = y.

Hint
This is exactly the inverse function theorem with F=sin⁡F = \sin, [a,b]=[−12π,12π][a, b] = [-\tfrac12\pi, \tfrac12\pi].

Solution

Apply the theorem of §110 with F(x)=sin⁡xF(x) = \sin x on [−12π,12π][-\tfrac12\pi, \tfrac12\pi], A=−1A = -1, B=1B = 1. It gives a unique function ϕ\phi on [−1,1][-1, 1] with sin⁡ϕ(y)=y\sin\phi(y) = y and −12π≤ϕ(y)≤12π-\tfrac12\pi \le \phi(y) \le \tfrac12\pi, continuous and strictly increasing from −12π-\tfrac12\pi to 12π\tfrac12\pi. This is arcsin⁡y\arcsin y, the value of the many-valued “arc sin” that lies in [−12π,12π][-\tfrac12\pi, \tfrac12\pi]. Among all solutions of sin⁡x=y\sin x = y, it is the one that vanishes with yy. (That sin⁡\sin is strictly increasing on this interval follows from sin⁡x2−sin⁡x1=2cos⁡12(x1+x2)sin⁡12(x2−x1)\sin x_2 - \sin x_1 = 2\cos\tfrac12(x_1 + x_2)\sin\tfrac12(x_2 - x_1), both factors being positive.)

Check yourself

Check yourself Choose one

Near (0,0)(0, 0), which equation does the theorem of §109 show defines a unique continuous y=ϕ(x)y = \phi(x) with ϕ(0)=0\phi(0) = 0?

Why

y3+y−xy^3 + y - x is strictly increasing in yy for every xx (a sum of two strictly increasing functions), continuous, and 00 at the origin. y2−xy^2 - x and y2−x2y^2 - x^2 are not monotone in yy near 00: the first has two solutions or none, the second has the two solutions y=±xy = \pm x, both continuous. x2+y2=0x^2 + y^2 = 0 holds only at the origin.

Check yourself True or false?

If FF is continuous and strictly increasing on [a,b][a, b], its inverse is automatically continuous: no extra hypothesis is needed.

Why

True: that is part of the theorem of §110. The inverse is strictly increasing and takes every value in [a,b][a, b], and a monotone function with no gaps in its values cannot jump (§101).

Check yourself Enter a number

How many real solutions yy does y5−y=0y^5 - y = 0 have? (This is the level x=0x = 0 in the reflection widget with a wide interval.)

Why

y5−y=y(y2−1)(y2+1)y^5 - y = y(y^2 - 1)(y^2 + 1), with real roots −1-1, 00, 11. So on a wide interval the level x=0x = 0 is taken three times, and x=y5−yx = y^5 - y has no single inverse; only on an interval where it is monotone, such as [−0.6,0.6][-0.6, 0.6], is the solution through (0,0)(0, 0) unique.

Hardy’s language

There is nothing to show that the ϵ1\epsilon_1 of the conclusion is the ϵ\epsilon of the hypotheses, and indeed this is generally untrue.

— G. H. Hardy, §109

Summary

Summary
  • If ff is continuous near (a,b)(a, b), f(a,b)=0f(a, b) = 0, and ff is strictly monotone in yy, then f(x,y)=0f(x, y) = 0 defines exactly one function y=ϕ(x)y = \phi(x) near aa with ϕ(a)=b\phi(a) = b, and it is continuous.
  • Proof: opposite signs at the top and bottom of a square persist sideways; on each vertical the intermediate value theorem gives a zero and strictness makes it unique; shrinking the square gives continuity.
  • A continuous strictly monotone FF on [a,b][a, b] has a continuous strictly monotone inverse on [F(a),F(b)][F(a), F(b)]: its graph is the reflection of FF’s graph in y=xy = x.
  • Without monotonicity the conclusion fails: y2=xy^2 = x has two solutions or none near 00.